A single stage multi-scale feature fusion convolution network integrated with vision based metrological characterization of cracks in concrete pavements is established to deliver an end-to-end crack detection and measurement (CDM) system. The convolution network considers the learning adaptability of the neurons and forces the neural model to evade correlation dependencies in learning features, resulting in a well-represented model. The well-learnt model exhibits high performance on crack induced structures. Post crack detection vision-based processing on the region encapsulating the crack is performed enabling extraction of crack edges by extending the Canny algorithm. Low mean error in estimated crack width validates the results. Finally, the integration of crack detection with its measurement provides a fully functional and implementable CDM solution.